Hello, I would like to ask how to select hyperparameters such as the number of neighbors and batchsize of Graphsage when comparing them to ensure that they have basically the same training conditions.